Dynamic image is encoded using prediction between images of the dynamic image in order to reduce redundancy in a time direction. However, in a fade-in or fade-out portion of the dynamic image, a difference value between images becomes larger than that in normal dynamic images, and it is difficult to reduce the redundancy in the time direction, and consequently, the encoding efficiency is lowered, the picture quality is degraded and the amount of generated code is increased.
In order to keep the picture quality of the dynamic images to be constant, following methods have been proposed.
(1) A method in which in a fade-in portion, encoding is performed by only in-picture prediction (see, for example, Japanese Patent Application Publication (KOKAI) JP-A-11-177989).
(2) A method in which a GOP structure of M=3 is changed into a GOP structure of M=2 (see, for example, Japanese Patent Application Publication (KOKAI) JP-A-2002-84544).
However, in the method (1), since the fade-in portion is encoded by only the intra prediction in order to keep the picture quality, the redundancy in the time direction can not be reduced, and in order to keep the picture quality, the amount of the generated code is increased.
In the method (2), when the fade-in portion is encoded, with respect to a B picture, distances to temporally forward and backward reference images become constant, and with respect to a linear fade image, efficient prediction becomes possible by bidirectional prediction. However, in a fade picture in which a luminance change is nonlinear, an improvement of prediction efficiency can not be expected, and with respect to a P picture, an improvement of encoding efficiency in a fade image can not be expected.